hex.grid

Class SequentialWalker<MP extends Model.Parameters>

java.lang.Object

hex.grid.SequentialWalker<MP>

All Implemented Interfaces:

HyperSpaceWalker<MP,HyperSpaceSearchCriteria.SequentialSearchCriteria>

public class SequentialWalker<MP extends Model.Parameters>
extends java.lang.Object
implements HyperSpaceWalker<MP,HyperSpaceSearchCriteria.SequentialSearchCriteria>

Nested Class Summary

Nested classes/interfaces inherited from interface hex.grid.HyperSpaceWalker

HyperSpaceWalker.BaseWalker<MP extends Model.Parameters,C extends HyperSpaceSearchCriteria>, HyperSpaceWalker.CartesianWalker<MP extends Model.Parameters>, HyperSpaceWalker.HyperSpaceIterator<MP extends Model.Parameters>, HyperSpaceWalker.RandomDiscreteValueWalker<MP extends Model.Parameters>

Constructor Summary

Constructors

Constructor and Description
SequentialWalker(MP params, java.util.Map<java.lang.String,java.lang.Object[]> hyperParams, ModelParametersBuilderFactory<MP> paramsBuilderFactory, HyperSpaceSearchCriteria.SequentialSearchCriteria searchCriteria)
SequentialWalker(MP params, java.lang.Object[][] hyperParams, ModelParametersBuilderFactory<MP> paramsBuilderFactory, HyperSpaceSearchCriteria.SequentialSearchCriteria searchCriteria)

Method Summary

Modifier and Type	Method and Description
java.lang.String[]	getAllHyperParamNamesInSubspaces()
java.lang.String[]	getHyperParamNames() Returns hyper parameters names which are used for walking the hyper parameters space.
java.util.Map<java.lang.String,java.lang.Object[]>	getHyperParams()
long	getMaxHyperSpaceSize() Return estimated maximum size of hyperspace, not subject to any early stopping criteria.
ModelParametersBuilderFactory<MP>	getParametersBuilderFactory()
MP	getParams() Return initial model parameters for search.
HyperSpaceWalker.HyperSpaceIterator<MP>	iterator() Returns an iterator to traverse this hyper-space.
HyperSpaceSearchCriteria.SequentialSearchCriteria	search_criteria() Search criteria for the hyperparameter search including directives for how to search and when to stop the search.
boolean	stopEarly(Model model, ScoringInfo[] sk) Based on the last model, the given array of ScoringInfo, and our stopping criteria should we stop early?

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface hex.grid.HyperSpaceWalker

estimateGridWork, getAllHyperParamNames

Constructor Detail

SequentialWalker

public SequentialWalker(MP params,
                        java.lang.Object[][] hyperParams,
                        ModelParametersBuilderFactory<MP> paramsBuilderFactory,
                        HyperSpaceSearchCriteria.SequentialSearchCriteria searchCriteria)

SequentialWalker

public SequentialWalker(MP params,
                        java.util.Map<java.lang.String,java.lang.Object[]> hyperParams,
                        ModelParametersBuilderFactory<MP> paramsBuilderFactory,
                        HyperSpaceSearchCriteria.SequentialSearchCriteria searchCriteria)

Method Detail

search_criteria

public HyperSpaceSearchCriteria.SequentialSearchCriteria search_criteria()

Description copied from interface: HyperSpaceWalker

Search criteria for the hyperparameter search including directives for how to search and when to stop the search.

Specified by:

search_criteria in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

getHyperParamNames

public java.lang.String[] getHyperParamNames()

Description copied from interface: HyperSpaceWalker

Returns hyper parameters names which are used for walking the hyper parameters space. The names have to match the names of attributes in model parameters MP.

Specified by:

getHyperParamNames in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

Returns:

names of used hyper parameters

getAllHyperParamNamesInSubspaces

public java.lang.String[] getAllHyperParamNamesInSubspaces()

Specified by:

getAllHyperParamNamesInSubspaces in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

getHyperParams

public java.util.Map<java.lang.String,java.lang.Object[]> getHyperParams()

Specified by:

getHyperParams in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

getMaxHyperSpaceSize

public long getMaxHyperSpaceSize()

Description copied from interface: HyperSpaceWalker

Return estimated maximum size of hyperspace, not subject to any early stopping criteria. Can return -1 if estimate is not available.

Specified by:

getMaxHyperSpaceSize in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

Returns:

size of hyper space to explore

getParams

public MP getParams()

Description copied from interface: HyperSpaceWalker

Return initial model parameters for search.

Specified by:

getParams in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

Returns:

return model parameters

getParametersBuilderFactory

public ModelParametersBuilderFactory<MP> getParametersBuilderFactory()

Specified by:

getParametersBuilderFactory in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

stopEarly

public boolean stopEarly(Model model,
                         ScoringInfo[] sk)

Description copied from interface: HyperSpaceWalker

Based on the last model, the given array of ScoringInfo, and our stopping criteria should we stop early?

Specified by:

stopEarly in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

iterator

public HyperSpaceWalker.HyperSpaceIterator<MP> iterator()

Description copied from interface: HyperSpaceWalker

Returns an iterator to traverse this hyper-space.

Specified by:

iterator in interface HyperSpaceWalker<MP extends Model.Parameters,HyperSpaceSearchCriteria.SequentialSearchCriteria>

Returns:

an iterator